We know the familiar theorem for the change of variables involving random variables (see for example here, Theorem 4.1):

Suppose $T:(x,y) \mapsto (u,v)$ is a 1-1 mapping from some domain $D\subseteq \mathbb{R}^2$ to some range $R \subseteq \mathbb{R}^2$. Define the Jacobian $J$ as a function of $u$ and $v$ by $J(u,v) = \left| x_u y_v - x_vy_u\right|$. Assume the partials exist and are continuous. If $X,Y$ have joint probability density function $f_{X,Y}$, then the random variables $U,V$defined by $(U,V) =T(X,Y)$ are jointly continuous with joint probability density function $f_{U,V}$ given by $f_{U,V}(u,v) = f_{X,Y}(x(u,v), y(u,v))J(u,v)$ if $(u,v)\in R$ and $f_{U,V}(u,v) = 0$ otherwise.

Okay, now suppose $X,Y \sim \mathrm{Unif}[0,1]$ independent random variables, let $T:\mathbb{R}^2 \longrightarrow \mathbb{R}^2$, be given by $(x,y) \mapsto (x+y, x-y) =: (u,v)$. We have $D = [0,1]^2$, $R = [0,2]\times[-1,1]$ and $x(u,v) = \frac{u+v}{2}$ as well as $y(u,v) = \frac{u-v}{2}$. We obtain our Jacobian as $J(u,v) =\left| \frac{1}{2} \frac{-1}{2} - \frac{1}{2}\frac{1}{2}\right| = \left| \frac{-1}{4} - \frac{1}{4} \right| = \frac{1}{2}$. As clearly $f_{X,Y}(x,y) = f_X(x)f_Y(y) = (1)(1) = 1$ by independence, we get by our formula that $f_{U,V}(u,v) = \frac{1}{2}$ on $R$ and $0$ everywhere else. A simple integration gives us $\int_R f_{U,V}(u,v)\mathrm{d}u\mathrm{d}v = 2$ which is clearly a contradiction as p.d.f's integrate to $1$.

Can someone help me and tell me where I made a mistake? I feel like this is such an easy question, but I can't really see what I'm doing wrong... Any help will be much appreciated!


The joint pdf of $U$ and $V$ does not have support $R = [0,2]\times [-1,1]$ as you claim it does; the support is a square region where the square has vertices $(0,0)$, $(1,1)$, $(2,0)$ and $(1,-1)$. The area is $2$ and the pdf has value $\frac 12$ on the region.

| cite | improve this answer | |

$$f_{U,V}(u,v) = f_{X,Y}(\frac{u+v}{2},\frac{u-v}{2}).\frac{1}{2}$$ $$f_{X,Y}(x,y) = f_X{(x)}.f_Y{(y)}$$ $$ f_{U,V}(u,v) = \frac{1}{2}, 0\le u \le2, -1\le v\le1$$ $$ \left[\int_{-1}^{0}\int_{-u}^{2+u} + \int_{0}^{1}\int_{u}^{2-u}\right]\frac{1}{2}dvdu = 1; $$

It is indeed 1.

The mistake is in your limits, you took u to be from 0 to 2 and v from -1 to 1 and you got 2 two's and a pdf of one half resulting in 2.

| cite | improve this answer | |
  • $\begingroup$ What is W, Z? What are you integrating? My integral is surely 2, as f is constant 1/2 on a square of area 4. We get 4*1/2 = 2 not 1. $\endgroup$ – Marcel S Sep 4 '17 at 17:09
  • $\begingroup$ X and Y gets transformed, you may have to review deriving joint probabillity density methodology. $\endgroup$ – Satish Ramanathan Sep 4 '17 at 17:38
  • $\begingroup$ sorry, I changed the solution. I was interrupted in the middle while I was writing the solution. I apologize for being shamelessly confident on a wrong solution. $\endgroup$ – Satish Ramanathan Sep 4 '17 at 19:36
  • $\begingroup$ The calculations are incorrect starting from the support of $f_{U,V}$ which is not as stated, and the integral does not match the alleged support. $\endgroup$ – Dilip Sarwate Sep 4 '17 at 19:41
  • $\begingroup$ Now the calculations are correct and supports the support. Thanks for correcting $\endgroup$ – Satish Ramanathan Sep 4 '17 at 20:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.